Analyzing infant cries to detect autism shows potential, but definitive evidence is still lacking

A systematic review integrating the findings of studies looking to identify autistic infants by analyzing the way they cry found several consistent acoustic patterns, but a pooled meta-analysis did not find statistically significant evidence for differences in pitch. There were indications from individual studies that the fundamental frequency (the basic pitch of the voice) of autistic infants’ cries might be higher, but this finding was not supported when all studies were combined in the meta-analysis. The paper was published in the Journal of Autism and Developmental Disorders.

Autism spectrum disorder is a neurodevelopmental condition that affects how people communicate, interact with others, behave, and process sensory information. Its signs can sometimes appear very early in infancy, yet many children are only diagnosed in later preschool years (or even later). This delay matters because earlier identification can give children and families faster access to support during periods of rapid brain development.

Current diagnosis relies largely on developmental observation, clinical assessment, and standardized diagnostic tools, all of which require trained professionals and can be difficult to access in some communities. Researchers are therefore exploring new ways to detect autism-related differences earlier and more objectively. Promising approaches include eye tracking, brain imaging, electroencephalography, video analysis, and other physiological or behavioral signals.

One particularly intriguing possibility is the analysis of infant crying. This approach looks into acoustic characteristics of an infant’s cry. Studies have reported that infants later diagnosed as autistic may differ in features of their cries, including pitch, duration, and variability. Because crying is universal, non-invasive, and easy to record, it could become a useful source of early biomarkers that complement existing autism screening methods.

Study author Sandra Pusil and her colleagues analyzed the results of studies that tried to use the analysis of infant crying to detect autistic infants. They note that these studies generally follow one of two approaches. One approach is to retrospectively compare acoustic features of cries of infants known to be autistic and those that are known to not be autistic. Studies following the other, prospective, approach follow infants who are considered to have an elevated likelihood of being autistic and compare them to infants with a decreased likelihood of being autistic. They then analyze their cries and follow them longitudinally examining which of them actually get diagnosed with autism later.

These authors conducted a systematic review and a meta-analysis of existing evidence. They conducted a comprehensive search of scientific publication databases that cover key biomedical, psychological, and multidisciplinary research areas, including PubMed, Scopus, Web of Science, and PsycINFO. They used keywords and subject headings related to cry analysis and autism in their search.

To be included in this review, studies had to focus on cry analyses, be conducted on children under 60 months of age, and either follow a uniform diagnostic protocol (for diagnosing autism), have diagnosis made by a specialist in neurology, psychiatry, or pediatrics, or be based on extensive, rigorous and detailed neuropsychological evaluation. The studies also needed to be reported in full text and available in English.

After searching the databases and removing duplicates, the records were screened. After excluding studies that did not contain what the authors were looking for, or that lacked the required data, they ended up with a total of 11 studies that were included in the systematic review.

Study authors rated 4 of these studies as being of good quality, while 7 were rated as fair. They included a total of 736 participants. The studies analyzed cry features of children ranging from neonates to 4 years of age. 5 of these studies followed the retrospective approach, while 6 followed the prospective approach. 6 of these studies were included in the meta-analysis, while the other 5 were excluded because they did not report the required data.

Multiple studies employed machine learning models to distinguish between the cries of autistic and non-autistic infants, consistently outperforming traditional statistical methods and often reporting accuracies exceeding 90%. Overall, some of the individual studies reported differences between autistic and non-autistic infants in the fundamental frequency of their cries, in the frequency of unusually high-pitched and intense cries (hyperphonation), voice quality irregularities, and other differences.

The most often reported difference was in the fundamental frequency of cries with multiple studies reporting higher pitched cries in autistic infants. However, when all studies were considered together, the difference between autistic and non-autistic infants in cry pitch (i.e., the fundamental frequency of cries) was not statistically significant (not sufficiently large to allow researchers to be sufficiently certain that it is not just caused by random variations).

“Cry analysis offers an objective, accessible, and non-invasive screening tool with the potential to support health assessments and developmental outcome prediction in early infancy. To realize this potential, future research must focus on large-scale, multicentric, longitudinal studies encompassing diverse populations. These efforts should aim to refine ML [machine learning] models, establish standardized datasets, and delineate developmental trajectories of acoustic markers,” study authors concluded.

The study contributes to the development of screening techniques for the early diagnosis of autism. However, it should be noted that most of the studies included in this review had very small sample sizes, requiring any effects to be quite substantial to be detected through statistical procedures used. Additionally, the studies differed very much in the age of participants, autism characteristics, and ways infants’ crying was elicited, further complicating the comparison of their results.

The paper, “Early Identification of Autism Using Cry Analysis: A Systematic Review and Meta‑analysis of Retrospective and Prospective Studies,” was authored by Sandra Pusil, Ana Laguna, Brenda Chino, Jonathan Adrián Zegarra, and Silvia Orlandi.